AUTHOR=Whiteley Louise , Sahani Maneesh TITLE=Attention in a Bayesian Framework JOURNAL=Frontiers in Human Neuroscience VOLUME=Volume 6 - 2012 YEAR=2012 URL=https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2012.00100 DOI=10.3389/fnhum.2012.00100 ISSN=1662-5161 ABSTRACT=A cornerstone of cognitive science is the idea that a process called `attention' dictates which incoming information is fully processed by a limited neural resource. However, attempts to discover exactly why the brain needs to selectively filter its input, and what the mechanisms and effects this of selection are, have floundered in a sea of heterogenous effects. This has led to assertions that a single neural resource allocated by attention is not a useful concept (Driver 2001; Zelinksy, 2005). By way of introduction, we briefly review the behavioural, physiological, and theoretical results that support this assertion, highlighting two different themes of attention research, and some of the debate that has gathered around them. In the main body of the paper we present a probabilistic framework under which apparently disparate resource limitations and attentional effects might be unified at the computational level. We exploit the idea of perception as Bayesian inference, and identify a general limitation in the brain's ability to perform ideal inference over the multitude of features present in a complex scene. We then suggest that attention acts as an adaptable Bayesian hypothesis to locally improve the impoverished stimulus representations that result. Finally, we illustrate this framework by modelling two key groups of attentional effects in a simple, generic setting.